Welcome to Data Science Dojo’s weekly newsletter, “The Data-Driven Dispatch”.
Some would argue that Artificial General Intelligence – machines matching human intellectual capabilities – is the doomsday of humanity. The thought of it reminds them of a robot apocalypse. On the flip side, optimists see AGI as the gateway to a utopian world where robots handle the grind, leaving humans to kick back and enjoy life.
But, let’s put this debate on the backseat for a second and let’s focus on the main question:
Are we ever going to achieve AGI?
Well, we could hope for it since we’ve already started to see the sparks of artificial general intelligence in OpenAI‘s recent multimodal LLM GPT-4.
One of the primary qualities that GPT-4 exhibits of AGI is common sense, a trait that sets humans apart. Here’s an example where GPT-4 demonstrates its ability to apply common sense, while ChatGPT struggles to do so.
Let’s dig deeper into what’s AGI, why multimodal LLMs like GPT-4 are pre-mature sparks of AGI, and what will be the implications of AGI.
Hang on tight as we get on this bumpy ride for you.
What is Artificial General Intelligence AGI
Ever heard of AGI? It’s like the superhero of AI, rocking intelligence that can go toe-to-toe or even outshine us humans.
These systems excel in diverse domains, seamlessly handling reasoning, planning, experiential learning, and natural language comprehension.
Unlike narrow AI systems that are designed for specific tasks, AGI systems possess general intelligence and can adapt to new and unfamiliar situations.
How are Multimodals like GPT-4 Early Sparks of AGI?
The 6 primary factors that would render a machine to meet the requirements of an AGI include the ability to reason, plan, solve novel problems, think abstractly, comprehend complex ideas, and learn quickly from experience. Interestingly, GPT-4 thrives in all these areas except for its ability to plan.
If you want to see GPT-4’s general intelligence in action, head on to this interesting read and explore its vast use cases: GPT-4: A Potential Stepping Stone on the Path to Artificial General Intelligence AGI
Then, the question proceeds to what makes GPT-4 achieve general intelligence.
We’ve listed down the primary factors of GPT-4 training as a model that allows it to have general intelligence, unlike other contemporary AI models which require special prompting to assist in specific tasks.
The Interdisciplinary Composition of GPT-4: GPT-4’s innovative design isn’t just a jumble of words—it’s a fusion of insights from diverse fields. From math and coding to vision, medicine, law, and psychology, GPT-4 seamlessly integrates knowledge from various domains.
This interdisciplinary synergy turbocharges its general intelligence, opening up a vast landscape of applications.
Multimodal Information: GPT-4 has the ability to manipulate and understand multi-modal information. It takes the cake in generating code that translates into clear and recognizable images, showcasing its grasp of visual concepts.
Picture this: you toss in an image of the website you fancy, and boom! GPT-4 whips up the entire code lightning-fast. And that’s just the tip of the iceberg How good is GPT-4 Multimodality
These features primarily contribute to why GPT-4 exhibits traits associated with intelligence, such as abstraction, comprehension, and understanding of human motives and emotions.
After GPT-4, Q* is Rumored to be the Next Breakthrough towards AGI by OpenAI:
OpenAI has not yet revealed Q* officially, but substantial hype has built around the reports provided by Reuters and The Information. According to these reports, Q* might have achieved AGI and is capable of solving problems that are beyond the reach of current AI systems.
Q* is reported to be a combination of multiple algorithms, including reinforcement learning, natural language processing, and machine learning. It is said to be able to learn and adapt to new situations at an unprecedented rate.
Read: Unraveling the Mystery of Q*: A Leap Towards Artificial General Intelligence AGI
Here’s a comprehensive lecture by Sebastien Bubeckwhere he assesses GPT-4 on the grounds of AGI, emphasizing its strong reasoning, problem-solving, and abstract thinking abilities but noting deficiencies in planning and memory.
He uses a 1990s psychologist-defined intelligence framework and concludes that while GPT-4 isn’t fully intelligent, it’s a powerful tool with transformative potential, urging caution in its usage. The video prompts contemplation on intelligence and GPT-4’s capabilities within the presenter’s chosen framework.
To connect with LLM and Data Science Professionals, join our discord server.
Can you relate with the level of happiness this one line brings🐵? Let us know in the comments.
We’ve seen that LLMs are slowly progressing to achieve general intelligence. Imagine what the future will hold if we keep moving at this pace.
Let’s meet this pace and learn more about large language models. Join Data Science Dojo’s Large Language Models Bootcamp– an extensive training that will enable you to build custom LLM-powered applications from scratch.
Want to learn more about the curriculum and instructors? Visit our webpage.
Finally, let’s wrap up this week’s discussion with some major highlights.
- Sam Altman was fired after researchers issued caution on a groundbreaking AI discovery called Q*.
- Amazon aims to make 2 million people ‘AI Ready’ by 2025 with new educational initiatives including 8 free courses on generative AI. Read more
- Microsoft’s Orca 2 elevates small language models with 7B and 13B parameters to surpass larger models in advanced reasoning. Read more
- Voltage Park secures $500M funding, and plans to deploy 24,000 Nvidia H100s for affordable AI compute leasing. Read more
- OpenAI And Microsoft sued nonfiction writers for alleged ‘rampant theft’ of authors’ works.
